Predicting Modified Fournier Index by Using Artificial Neural Network in Central Europe.

Area of Science:

• Environmental Science
• Hydrology
• Artificial Intelligence

Background:

• The Modified Fournier Index (MFI) quantifies rainfall erosivity, a critical factor in soil erosion.
• Predicting MFI is essential for effective land and water resource management.
• The application of artificial neural networks (ANNs) for MFI prediction remains underexplored.

Purpose of the Study:

• To evaluate the performance of Multilayer Perceptron (MLP) and Radial Basis Function (RBF) ANNs in predicting MFI.
• To identify optimal climate data scenarios for MFI prediction using ANNs.
• To assess the relative importance of climate variables in MFI prediction.

Main Methods:

• Collected climate data (precipitation, temperature) from three Hungarian stations (1901-2020).
• Calculated MFI and trained MLP and RBF neural networks under four different input data scenarios.
• Evaluated prediction accuracy using Nash-Sutcliffe Efficiency (NSE) and correlation coefficients.

Main Results:

• ANN models demonstrated good to accurate MFI prediction capabilities (NSE values ranging from 0.68 to 0.73).
• Correlation coefficients between observed and predicted MFI were high (0.83–0.86).
• Scenario SC2 (P + p) and SC4 (P + T + T + T) were identified as optimal for MLP and RBF, respectively, with precipitation and temperature being key predictors.

Conclusions:

• ANNs, including MLP and RBF, are effective tools for predicting MFI in Central Europe.
• Climate data, particularly precipitation (P, p) and temperature (T), significantly influence MFI prediction.
• This research provides a foundation for utilizing AI in rainfall erosivity assessment and soil conservation strategies.